| name | features-rewrite |
| description | Multi-agent convergence loop for rewriting service features.md. Spawns Opus writer + 6 critic agents, iterates until quality threshold met. Use when rewriting features docs or running the convergence review loop. |
| argument-hint | <service-name> | --list |
Features Rewrite — Multi-Agent Convergence Loop
Rewrite a service's features.md using 1 Opus writer + 6 critic agents iterating until convergence.
Usage
/features-rewrite <service-name> # Rewrite a specific service's features.md
/features-rewrite overview # Rewrite docs/overview.md with full 6-critic loop
/features-rewrite --list # Show all services with features.md + last modified
Core Mandates
- Never write without vision input — Extract from
memory/feature-review-log.md or ask the user
technical.md is the sole fact source — Every claim must be verifiable against it- Maximum 3 rounds — Diminishing returns proven over 24 rewrites
- User approves final output — Never overwrite features.md without explicit approval
- Log every run — Append scores, outcome, and key corrections to
docs/features-rewrite-history.md
- Check overview drift — After every service rewrite, validate
docs/overview.md for stale claims about that service
Invocation Detection
| User Says | Action |
|---|
/features-rewrite <name> | Execute workflows/rewrite-single.md for <name> |
/features-rewrite overview | Execute workflows/rewrite-overview.md |
/features-rewrite --list | List all docs/services/*/features.md with stat dates |
| "rewrite features for X" | Execute workflows/rewrite-single.md for X |
| "rewrite the overview" | Execute workflows/rewrite-overview.md |
| "run the convergence loop on X" | Execute workflows/rewrite-single.md for X |
Pre-Flight Checks
Before starting, verify all 3 inputs exist:
| Input | Path | If Missing |
|---|
| Current features.md | docs/services/<name>/features.md | STOP — service may not exist |
| Technical reference | docs/services/<name>/technical.md | STOP — no fact source |
| Owner's vision | memory/feature-review-log.md entry | Ask user via AskUserQuestion |
Agent Architecture
| Role | Model | Count | Purpose |
|---|
| Writer | Opus | 1 | NYT-quality prose, structure template adherence |
| Platform Architect (PA) | Opus | 1 | Technical accuracy, cross-service correctness |
| External Developer | Sonnet | 1 | Zero-context comprehensibility |
| Product Strategist | Sonnet | 1 | Value proposition, differentiation |
| End User | Sonnet | 1 | Non-technical accessibility, jargon detection |
| First-Time Builder | Sonnet | 1 | Architecture clarity, inspiration |
| Competitor Analyst | Sonnet | 1 | Competitive positioning, moat communication |
User Checkpoints
| When | What User Sees | Options |
|---|
| After writer's draft (each round) | Full draft in features.md | "Proceed to critics" / "Redirect with feedback" |
| After final convergence | Final draft + scorecard summary | "Accept" / "Provide additional direction" |
| After overview drift check | Drift report for changed service | "Rewrite overview now" / "Defer" / "Ignore" |
Workflows
| Target | Workflow | Trigger |
|---|
| Single service | workflows/rewrite-single.md | /features-rewrite <service-name> |
| Platform overview | workflows/rewrite-overview.md | /features-rewrite overview or drift check → "Rewrite overview now" |
References
| File | Contains |
|---|
reference/writing-rules.md | NYT style, banned words, structure template |
reference/convergence-criteria.md | Scoring thresholds, decision matrix, max rounds |
reference/critic-roles.md | 6 critic personas and evaluation axes |
templates/writer-prompt.md | Writer agent prompt with placeholders |
templates/critic-*.md | 6 critic prompts with placeholders |
templates/scorecard.md | Structured output format for all critics |
workflows/rewrite-overview.md | Overview convergence loop (all 24 features.md as input) |
docs/features-rewrite-history.md | Rewrite audit trail — scores, outcomes, corrections per service |
